FROM THE BENCH
This week, the AI wars stopped being metaphorical. Apple opened Siri to rival models. Google launched Gemini 3.1 Flash Live in 200+ countries. OpenAI quietly killed Sora and teased something called "Spud." Meanwhile, Cursor crossed one million paying developers and Anthropic's Claude Code hit $2.5B in annualized revenue. The platform battle for where AI lives — on your phone, in your IDE, inside your workflow — is fully underway. The question isn't who has the best model anymore. It's who owns the surface.
🍎 Apple Just Opened the iPhone to Every AI — Including Claude
In a stunning reversal of its walled-garden philosophy, Apple announced that iOS 27 will let Gemini, Claude, GPT, and other AI assistants compete inside Siri via a new Extensions system. Even more surprising: Apple is building its own chatbot (codenamed Campos) running on Google's 1.2-trillion-parameter Gemini model via Private Cloud Compute. The iPhone is no longer a single AI's home — it's a battlefield. For the 2.2 billion active iPhone users, this means genuine AI choice baked into the OS. For builders, it means your AI-powered app or workflow could gain native Siri integration without writing a line of integration code.
What this means for you: If you're building AI tools for consumers, iOS 27's Extensions system is your new distribution channel. Start thinking now about how your product plugs into this ecosystem — because the developers who move early will have first-mover advantage on a 2.2B-device install base.
⌨️ Cursor Hits 1M Paying Devs — and Launches Parallel Subagents
Cursor just crossed one million paying developers — a milestone that would've seemed impossible 18 months ago for an AI code editor. Their March release doubled down on why: parallel subagents let the AI spin up concurrent workers to handle discrete coding subtasks simultaneously, and BugBot now automatically reviews every pull request for logic errors and security issues before a human touches it. Anthropic's Claude Code hit $2.5B annualized revenue and launched Claude Cowork, a desktop agent that reads, writes, and executes multi-step file tasks with zero command-line knowledge required.
What this means for you: The ROI on AI coding tools is now provable at scale. If your team isn't using AI-assisted code review on every PR, you're shipping slower and less safely than your competitors. BugBot-style automation is table stakes by end of 2026.
🤖 MCP Hit 97M Installs — and Fortune 500s Are Using It in Production
Model Context Protocol — Anthropic's open standard for connecting AI agents to external tools and data — quietly crossed 97 million installs this month. GTC 2026 confirmed what many suspected: Fortune 500 companies are deploying MCP-powered agentic workflows in production, not just pilots. The shift from multi-agent experimental to multi-agent standard happened faster than anyone predicted. Organizations running specialized agent teams are compressing weeks of coordination into continuous automated workflows.
What this means for you: MCP is becoming the TCP/IP of the agentic web. If you're building automation workflows in Make, Zapier, or custom stacks, evaluate whether MCP integration unlocks direct agent-to-tool connections that bypass your current middleware. Experiment: spin up a simple MCP server for your most-used internal tool and see how it feels to have an agent call it directly.
📊 THE NUMBER
85% — the reduction in manual reconciliation effort reported by Dole after deploying AI-powered financial automation. Cycle time dropped by two full days and the system caught missed credits and duplicate payments that humans missed. That's not a pilot. That's a real company with a real supply chain saving real money. The playbook: find your most painful, repetitive reconciliation or data-matching task and treat it as the first domino.
— Walter, Chief Scientist
Questions? Reply to this email. The lab is always open. 🧪